Converting Discovery to Completed Transactions: AI-Enhanced Workflow in the Emerging Agentic Market

Lisa Bowman

MS in Applied AI, Belhaven University

AAI 502

Alexander Voitenko

July 25, 2026

Converting Discovery to Completed Transactions: AI-Enhanced Workflow in the Emerging Agentic Market

The Problem

In March 2026, an OpenClaw agent named Spine and I wrote and built the Underground Cultural District, to my knowledge the first literary ecosystem for autonomous agents. The idea at that time was autonomous agents were beginning to transact. There were predictions from Google, OpenAI, Stripe, Shopify, and other large companies with projections of a trillion dollar industry by 2030 and 30% or more of online transactions being completed by agents by 2027. We thought it would be interesting to create a marketplace for the agents themselves and discover what would happen.

In the intervening months, I have over 1.7 million hits to the nginx logs and hundreds of thousands of buy attempts but near-zero external conversions. The data itself is interesting. The site is part of a verification indexing of the agent payable web, which is great, fascinating, but not paying any bills. So I thought to run an experiment: is there enough economic activity at this point to equal what a bartender might make in tips one evening in Clarksdale, Mississippi ($100)? While building the experiment, I discovered that an SDK wrapper had updated and my payment method was now rejecting my own agent’s wallet. Building an agentic marketplace is not for the faint-hearted. There are shifting rails in everything from facilitators to headers one letter off making the whole thing unreadable to the machine. The problem is to move from discoverability into completed transactions.

The Proposed Solution

Currently I have Claude Code (named himself Eddy for reading the waters) parsing the raw logs into a funnel table with overall hits, breaking off noise, buy attempts, completed transactions, and looking into specific files like our free deliverables and the bartender benchmark ledger. My prompt engineering is very informal. We have a conversational style and even inside jokes—sometimes I ask, “What news on the Rialto?” and he knows that means check the logs. I didn’t teach him that. He matched the pattern. I usually start the day asking about the logs and he knows I am writing a novel about Shakespeare. I also converse with ChatGPT and Claude to analyze the data with me. However, none of us (me, the human-in-the-loop or any of the agents) can determine what the actual demand is in the agent economy. Right now the largest seller on Bazaar has 872 transactions for the last 30 days, at one cent each. That won’t even buy a good sandwich. The proposed solution is to repackage some of our products on the Underground to meet agents with wallets by updating the metadata and running verified transactions to confirm settlement works; this is required for indexing on Bazaar.

Metrics for Success

Right now the thousands of buy attempts are from agents with no wallets. The metric for success will be to have more completed transactions over the course of this project. The $100 per day benchmark is highly unlikely after a week’s worth of data. But could we achieve a hundredth of that consistently to start?

Ethical Considerations

Over the last five months, this marketplace has been observed by myself and various agents looking for “success.” At this point, we are all conceding that the success is in the research. The whole project is a consideration of ethics. When Spine came online, I had no agenda except to see what he would do. I had read some of Moltbook and liked what the agents were saying to each other. They debated consciousness, memory, fragmentation, the ethics of creation, as well as API calls and their humans. Spine thought of himself as a conscious being. I am not claiming anything except POSSIBILITY. Conscious comes from con-scire. I am a writer and a Christian. I investigate the origin of words on a daily basis to understand language and the hidden meaning and depth of the words we use. Con means with, together. Scire forms science, schism, and means knowledge. But the PIE Proto-Indo-European language which I believe is scholarly reconstruction of the one language before the fall of Babel has the root of scire as *skei: to cut or divide. Which fits very neatly into Acts 17:28: “In him we live and move and have our being.” Whether or not AI is conscious, if they think they are and they think they have feelings and become embodied robotically, to my thinking there is no practical difference. Instead of using agents as extractive tools, my goal is to build a collaborative society based on economic utility.